The Doctor Policymaker

Right — I blame my parents. I'm Simon Eccles, National Chief Clinical Information Officer for Health and Care, a role based in NHSX but covering NHS England, the arm's-length bodies and the Department of Health and Social Care. I'm also Deputy Chief Exec at NHSX, the NHS's digital innovation arm. And occasionally I'm still an A&E consultant at St Thomas' Hospital. How did I get there? I did the usual medical school stuff.

It's probably worth telling this bit. It started when I was a house officer. I was asked to consent parents of children with cleft lip and palate for surgery. I'd never seen a cleft lip and palate operation, let alone knew what the complications were. I asked, and got told, sort of, "just do it." And I refused — which was slightly brave — and I ended up in front of the Dean, who said, actually, that was probably fair enough, and I ought to at least know what the operation consisted of before consenting parents. As you probably know, you now don't consent unless you're competent to undertake the procedure.

But that got me involved in medical politics, and through medical politics, in trying to make stuff better. Wind the clock forward a while: I was a senior registrar, and the opportunity to get involved in digital work became clear. It occurred to me that if we could do what was then "IT in healthcare" better — join records up, so medications became known and allergies became known — we would make care safer, make doctors' and nurses' lives easier, and involve patients in their own care better. I thought that was phenomenally worthwhile. And slightly naively, I thought it wouldn't take terribly long to do. It's now 14 years later — I've gone out of it and come back into it — and I'm still trying to do more or less the same thing.

There are two — I know you asked for one, so I'm being generous. At a senior level in an organisation, staff are familiar with how long it takes to make change. At a junior level it seems easier — and the reason it seems easier is that you can make a change, and then you leave: it's a six-month or one-year rotation. What you don't see, after you've gone, is that the idea probably peters out. Something that actually sticks, and becomes a complete change in the way work is delivered, is really hard, even within a single NHS organisation. Doing that across 200 acutes — or 7,000, if you're doing general practice — is extremely challenging. What you need is a compelling reason for change.

Recently, with coronavirus, we had an absolutely compelling reason to change: to move from face-to-face consultations by default to telemedicine, online or telephone consultations by default. And the NHS did it. It just did. We put virtual consultation into virtually every outpatient and primary care setting, and that change was wholesale and absolutely transformational. What's going to be interesting is, as coronavirus eases as a pressure — certainly next year, even if not this winter — how do we maintain the good things about that change? I'm not getting into some battle about whether every consultation should be remote. No, they shouldn't all be remote — but more probably can be. And that should largely rest with a combination of the doctor and the citizen making a decision between them about what suits that consultation best. So that's the first misconception: it's harder than you think without a compelling reason for change.

The second misconception is the scale of the NHS. It's huge. We burn through — whatever it is — £140 billion a year, amounting to about eight and a half or nine percent of GDP. As a consequence, sums of money that seem really large can just disappear into the NHS. We say "we're spending £50 million" and you go, "wow, £50 million quid" — right, that's £250,000 per acute hospital. And if that's over a year, for an acute hospital with a turnover of £300–400 million, it's not going to make a phenomenal difference. So when we look at changes like implementing electronic patient record systems — which are of the order of £30 million for a small organisation, ranging up to about £100 million for a big one, times 200 — you end up with sums of money that even the Treasury goes pale and weak at the knees over. So it's not as surprising as it might be that this stuff takes time, is complex, and costs more money than people expect. That does make me sound defensive, but live with that.

Oh God, you're into some heavily existential questions there. And the answer is both. I think we do too much one-size-fits-all for some things where the actual change is inevitably delivered locally. A change in the way we deliver care is going to end up with a clinician, a nurse, an AHP and a patient having a different interpersonal relationship — whether that's done remotely, or because they've both seen the same set of blood results, or a co-produced care plan, or whatever. Right down there, it's local. So a central edict isn't necessarily going to influence the millions of individual patient interactions.

At the same time, we are one NHS. Almost every patient I've ever spoken to about this can't quite understand why the record of their past operations, their current medications and their previous illnesses isn't known to any point of contact they have in the NHS — whether that's the ambulance staff turning up at their door, the A&E department, or an out-of-hours GP. Why haven't they got my records? It's all part of the NHS. So there's some stuff we really ought to do better nationally, about setting standards and ensuring availability. But when it comes to transformation and change, the unit of currency is as local as it gets.

Let's come back to the private–public piece in a moment, if we may, because that's a whole different ballgame.

But electronic records are a must. We were one of the first nations to fully digitise elements of the record. All primary care records — all GP records — are digital across the NHS. Technically maybe one or two practices are still on paper, just for fun, but everyone else is digital. We were one of the first to have fully digital imaging across the whole country, and most pathology is processed digitally — even though some is still, extraordinarily, ordered on paper. So we've come a long way. All the national systems are now in place in every organisation, and our data is collected, by and large, digitally.

But if you're working in a hospital that either doesn't really have an EPR system — and there are a few — or that has lots of individual bits of systems that don't talk to each other, and you've got a Post-it note stuck on the back of your ID badge with 12 different usernames and passwords, each going out of date at completely irregular intervals... it doesn't feel like you've got a coordinated system. And yet we've got some of the best digital hospitals in the world in the UK — I'd include Cambridge, West Suffolk and some others at the highest level of digitisation.

So our challenge is: how do we raise the general expectation for digital records, so that it becomes unacceptable — from a safety perspective and an efficiency perspective — for a hospital's records to be anything other than digital? And how do we help hospitals get to that point? My wishing it to be true doesn't make it happen. But it should be no more acceptable for a junior doctor to be writing on blood bottles by hand — hoping they've spelled the name right and got the date of birth correct — than it is to have the roof of an operating theatre leaking. Those are safety concerns; they shouldn't happen.

You then asked how we're going to make it happen. With the National Programme for IT, we spent a lot of money — see-it-from-space amounts of money — trying to make this happen, and it didn't work as successfully as it might have done. So we need to move to a situation where coordinated electronic patient records are expected in every organisation delivering care, including secondary care, and where we've got a funding model that reflects the complexity and the need to train staff in how to use them.

The second challenge is: it's all very well everybody being on their own digital island, but how do we join all of those together? As a nurse or doctor receiving someone into A&E or onto a ward, or as a GP whose patient has just been discharged, how do I absorb the right information from their previous care and make it part of what I'm doing now? And I want to go further than many others worldwide have gone.

There are two models live on this at the moment. There's the US model — probably best exemplified by Boston — a universal record, but it's a passive view of others. You create your own record; you can see other people's and cut and paste from theirs. So the record slowly accumulates volume as people grab bits and add their own, and it gets pretty huge. Or you've got the Hong Kong model — a single record system nationally across the island, where everybody's health records are on it and every point of care can see it.

We're going to go for neither, because we've got independent organisations: GPs on their record systems, mental health on its, community care on its, and the hospitals on different systems. I don't think one system for the NHS is the right idea. I've worked with single vendors for solutions in the past, and if anything goes wrong, it goes wrong for everybody at the same time. That's really bad — you get an outage, and the NHS goes offline for a day. And if people don't like it, they've got no choice. So I think it's actually a good idea that there are different systems out there, that some are thought to be better than others — and we need to get much better at listening to the users.

The vast majority of keystrokes are made by nursing colleagues, and second to that junior doctors. Those users need to be saying, "this is good, and this is why," and "this is rubbish, and this is why" — and the developers need to hear that feedback. What they tend to get is feedback from the IT departments and the informatics professionals, not necessarily from the front line.

The other element is aiming for the concept of a single source of truth — how the modern internet is structured. Take two examples. If you've come into hospital and I want to know your last full blood count, I don't want a copy of it sent to me. I want the single source of truth of your full blood count. Because if somebody realises there's an error — the analyser was wrong, the numbers were wrong, the bottle was mislabelled — I don't want an out-of-date copy. I want to know what was real at that time.

And if we take medications: at any given moment, a citizen has one medication list. It may include medications from their GP and from their specialist. Many patients have depot injections once a month that often get forgotten on medication lists, because they don't occur to people. We need that full list.

And if we're going to go further — and I hope we will — I want to know the indication for those medications: the start indication, and the stop indication. Antibiotics or analgesics come with relatively clear stop indications — after the pain's gone or the condition's better, they stop. Some, like anti-epileptics, you look at stopping or tapering after someone's fit-free for six months or a year. Antidepressants may be weaned or changed — with patient consent, and with those indications made clear. Others have regularly changing dosages, dependent on blood test results. That becomes much clearer to a clinician who hasn't met the patient before — and indeed to the patient themselves — why the dose changes, what we're looking for, and how often it ought to be measured. So we can get a much richer picture about medicines than a single list.

I also want to know what medicines have been tried in the past and why they were discontinued for a given individual. We see this in my field of emergency care, where someone comes in on a bunch of painkillers, their pain's still ongoing, and it's not working. "Gosh, everyone else has been treating you like a complete idiot — what I need to do is change this to a non-steroidal, and everything will be better." The patient goes back to the GP: "Oh, for God's sake, that's the one we tried six months ago — it gave you the dreadful drop-off in your kidney function. Bloody A&E, cross it off again." If we're sharing all of that information, we'll provide much better care. So EPRs need to be everywhere, and some aspects of the information they hold need to be common to all providers. That, I think, gets us much closer to a real value and safety case.

Blockchain. I don't want to diss an entire industry based on cryptocurrency and possibly money laundering, but blockchain is one of those technologies for which promises keep being made, some of which sound extremely compelling.

Take medication: the ability to use the blockchain to verify a medicine's veracity — that it contains the active ingredients and isn't fraudulent — from the point of manufacture, through each point of wholesale and resale, to the pharmacy, retail or hospital, to the ward or bedside, to use. That's compelling. And I've heard others around identity verification, where the blockchain may be part of the answer. But there isn't a use in practice, and it's been around for a long time. I think the hype curve has slightly disappeared for that one now — and with a bit of luck, as it calms down, we'll be able to see where the real use cases are. I've seen it in retail and in product tracking, but I'm not convinced it's got as widespread uses as others make out.

That's a great question. I'll try to come up with ones you haven't already mentioned — but there are some I'm surprised aren't making greater inroads. Natural language processing is the biggest. I'm old enough to be almost pathetically analogue — even as we talk, I've got a pen and paper nearby. When it comes to creating large volumes of words on a page, I'm not a touch-typist, so I dictate it all into an NLP program that works extremely fast and has learned my voice.

There's an A&E department in the North East that made that technology available to their staff, who can record notes more than twice as fast as they used to by typing. Almost all radiology reports now are created as a product of NLP-backed dictation. And yet it hasn't made it widely into the rest of the NHS — and we're talking about technology that's on every single smartphone, in the pocket of just about every doctor and nurse in the country. What I'm after is a secure means by which the device has no idea what words are going in and out of it — so we're not giving away patient confidentiality — but which can then be squirted into the clinical record as lumps of text.

The second order — which was shown to me over a decade ago — is the part of NLP that understands what a block of text means. You create a screen of clinical free text by dictation, and the device goes, "I think this is what you mean," structures it, codes it, and puts out those structured codes: "do these fit?" For an operative note, that would include diagnosis, procedures, laterality — left, right, both — and negation. "There was no evidence of cancer" — you don't want to just pick up the word "cancer" for coding.

I saw beautiful examples at the World AI forum in Shanghai that could take huge pages of handwritten text and fire them through. So the first piece is interpreting the script, to convert it into something machine-readable; the second is machine-reading it, pulling out the meaning, and structuring and coding it. If we get that right, we can start structuring and searching records much more usefully than we can now. Google does this routinely for pages of text. It's harder in medicine, I get that. But if I want to look at someone's past blood results, unless they've been formally coded, it's really difficult — and yet if you or I read their notes or clinic letters, whether handwritten in a bundle of paper or stored in any document system, we'll spot a blood result almost instantly. Three digits, dash, two digits, usually. I get excited if it's two-over-two or three-over-three; anything else doesn't matter — that's a joke — but I'm surprised we haven't adopted technology that makes that absolutely routine.

Second order: we've seen machine learning working well at image interpretation — the most obvious example. We need to be quite caveated here: it's not that the machines can read every X-ray better than humans. But give them a series of pretty defined parameters and they're very good at it. Lesion follow-up: we've got a known nodule on a CT scan — has it got bigger? The machine will do that volume calculation better, and much faster, than a human. Retinal scans: huge volumes are now done, both on the high street and in specialist clinics, and you want to understand whether they're normal or whether something needs attention. You can do that screen really quickly with AI. The AI is not bad at spotting what the human would — but I'd suggest you double-read it. What you can do is clear the normals, reduce the backlog, and make the task much easier. So I'm surprised we haven't done those.

The third is almost boring: the mass sharing of already-existing information and data. We still hold as our gold standard — rightly — evidence-based medicine: I need to see the controlled trial between whatever I'm interested in and the best alternative, to see whether it's better. But the real world, particularly of prescribing, isn't as straightforward as that. The patients aren't necessarily the same age as the group who volunteered for the trial; they're on polypharmacy; they're often not completely compliant with their medications. Is this going to work in the real world? Well, we hold that data. If I want to know, for my 88-year-old father, whether it's worth him taking a statin given the side effects he's getting from it, I want to know the benefit to 88-year-olds with a degree of chronic renal failure of a statin over the next ten years — and the NHS holds that data.

I'm surprised we aren't making better routine use of huge volumes of information. We are getting better over time. But 12, 14 years ago, Tesco was using machine learning to look at the queues in supermarkets, to redesign their checkouts and determine their staffing rotas to better reflect the numbers passing through. And they got good at it, because if the checkout queues are bad, people don't use that supermarket — they go to another one. The NHS occasionally behaves like Mondays are a surprise — on a weekly basis. The best organisations are brilliant at predicting bed demand, but others still treat each bed request as if it's a quest for the Holy Grail, rather than a completely predictable event that a pretty modest level of routine retail demand-prediction would forecast to as near as makes no difference.

I'm not going to give you any examples that aren't freely available, I'm afraid — but yes, there are lots doing fun things.

Estonia is always quoted as the most digital nation. They're not that big — 1.3 million or whatever it is — but they have fully digitally coordinated records for all public services. There's something about how the world would look if you knew all your records were always going to be available — the difference it would make to how we structure records, how we structure handovers, how citizens approach new points of care. They've thought about that very well. Who else would I rate? Australia is trying to do great things with remote telehealth — they've always been pretty good at it because of the geography — and they're spending on My Health Record at the moment, trying to share medications and other elements. The Dutch have adopted a completely different approach to the public–private partnership we were discussing: the public part is a very thin layer on top of a privately provided system. I spend my life talking to other countries, and everybody struggles in different ways with similar things. Interoperability is universally, internationally challenging, and no one's got it sussed.

But we have the Global Digital Health Partnership set up — I think we've got about 60 nations on board now — looking at interoperability, among other challenges, to ask: how do we help each other with international standards, so we're not individually wasting effort? SNOMED CT as the underpinning coding standard, FHIR probably as the underpinning messaging standard, HL7 as the structuring standard — these are all good. They are good. But I suspect that within not terribly long they'll be somewhat redundant, because if the AI gets better in the way we were discussing earlier, you won't need to be as structured to still be fully machine-readable. I don't know whether I'm being overly optimistic, but there's a bit of me that wonders if we're spending a long time describing the right standards for video between VHS and Betamax — and no one's quite spotted discs, and they sure as hell haven't spotted live streaming — and perhaps we're targeting the wrong problem. But nobody's got that answer yet. I keep watching. I'm promised a lot. Blockchain effects.

I did the Major Projects Leadership Academy at Saïd Business School, over 18 months to two years. It was fascinating. There are a number of bits I've used since, as they apply to all projects. We had an air vice-marshal in my group who was extremely funny about an aircraft type that has to remain nameless; the chap responsible for the British nuclear submarine fleet; people from all sorts of parts of HS2, and so on. But the "black swans", as they're called — the project which, as the organisation tries to pursue it, consumes all the resources, all the money, and then goes desperately wrong — they were all digital. Indeed, the National Programme for IT was used as the par excellence example, so I became the butt of everybody's jokes. But that was fine.

It was interesting, with a group of such interesting people at the top of their game, discussing why projects like that go wrong — that's relatively easily understood — but how they're allowed to keep going wrong, to the point where they're consuming huge amounts of resource and people have quite clearly lost their grip. One thing that struck me: if I took a finance spreadsheet for tens of millions on something hideously complicated to the board — whether it's the board of the CCG or of the acute hospital — they'd understand that spreadsheet. You could walk them through it: the benefits case, the ratios, what's capital, what's revenue. Fine, no problem. If I took a clinical safety case — "so-and-so died because they were given a medication they were allergic to, because nobody put the wristband on, because the ward couldn't find the wristband" — everyone would be able to read it, understand the root-cause analysis, and understand what we're proposing to do differently.

But you take an IT case to the board, and in many organisations you'll have that same conversation — while in some, you're really working uphill, and they'll look at the CIO or CCIO and hope they're competent enough to answer it. There isn't yet the same innate understanding that digital is a core part of how we transact business. If you go to a large retailer these days and the board doesn't understand digital, they're going to go bust — it's now the world in which they work. It needs to become the world in which we work, so that in close to real time we can understand what our data is telling us.

We've got the GIRFT programme — Getting It Right First Time — helping organisations understand their quality data and what to do to improve it. That's really interesting to see, but it becomes meaningful when you're looking at this month's or last month's data, not six months' or a year's ago. In one of the best organisations I've come across, a surgical unit employed a research registrar to look at their coding data — to help every junior doctor do their audits, to help their national coding, and to help them produce brilliant research. I was talking to her, and she said, "and then the doctors change over at the beginning of August, and I spot it" — she's like the Matrix, watching the screen in front of her: "Ah, you, Ward 7, you're doing it all wrong." And she'd be bleeping them at once: "No, no, you've completely misunderstood how to do this." That approach to data quality is just fabulous and revolutionary, and we need to get to a point where it's widely understood.

So I was there at Saïd Business School, listening to these really interesting people, and the way to run something well is to understand in really interesting detail where its risks are — not as an abstract risk register, but in a practical application of reality. The number of digital projects that have gone wrong because the top three things on their risk register came true is ludicrous.

The other bit I took from it — which lots of people have done — is about personality types and leadership styles. For me, there were two aspects. One: how each style of leadership goes wrong when carried into caricature. The person who loves order and structure and is methodical becomes very static if they go too far; the person who loves dynamism and drive becomes angry and ranty. And this idea of the complete leader as a perfect blend of all types — no, none of us are, it's simply not true. So how do you create a team that reflects all of those styles? How do you work with them as individuals — I've never used "empower" — to make sure all of our collected knowledge is fed into how to deliver? That's both the clinical knowledge, the digital knowledge and the transformation skills. Where you can do that, you stand an infinitely better chance of creating something truly useful to the service — rather than building something in isolation, dropping it over the fence, and wondering why nobody wants to pick it up and use it.

I think there are an awful lot that haven't. My inbox is a howling disaster — a source of much mirth to others. But there's a bit about people holding themselves back, and I'd suggest: don't.

A good idea is a good idea. People tend to be quite reticent, or the first person they discuss it with squashes it. I was advised pretty seriously, as a junior doctor, by more than one consultant, that my failure to only pursue the specialty training in hand and the next exam was a sign that I was completely hopeless — that this ridiculous interest in other things was a flash in the pan and should be stopped. I'm glad I didn't listen to them as much as I might have done. But it was not easy, and it was not made easy.

Now, I'm aware of a very small number of training programme directors who recognise that it may be okay to let someone train part-time to pursue an interest in digital technology alongside the other aspects of their time. And yet we desperately need clinical informaticians in the health service — we need to create a roadmap for them.

So if you're listening to this and it's a real interest for you: it's creating a slightly dogged personality that's going to get you through some of this, while we as a profession catch up. I'm committed to creating a career path, but until that's transparently obvious to others, you've kind of got to push through a bit. It is hard. Not giving up on that, I think, would be useful.

Other habits — I don't know. I write endless to-do lists. I'm not sure I'm any good at it; I find myself adding things to the bottom of to-do lists in order to cross them out, so there's something on the list that's been done. That's not a good habit. It's a dreadful habit, but it cheers me up.

I'm not going to give you any advice others haven't given a hundred times before, around looking after yourself and creating sufficient time away from work. Medicine can be all-consuming; IT can be all-consuming.

One more thing: get an honest, neutral opinion of whether the idea — the career idea — is a good one, and not just from your current boss. Who do you trust to give you that advice and guidance? I came across a couple of junior doctors who mortgaged their houses to create an app. It was a perfectly valid idea — let me be quite clear about that. But it was launching into a super-competitive field, and I could name eight others. I was saying, "hang on — you've mortgaged your house for this? It's nowhere near ready. And it's your first? Your first app's going to fail — that's the way it works. Your second one may be right; your third may be genius; and with a bit of luck your fourth is a whale and makes you a billionaire, and that's all marvellous, and I should be bloody jealous. But it isn't the first one." So I think independent sources of really good clinical-digital careers advice for doctors are very, very hard to come by.